Reflector: Arrangement-Aware Harmonic Retrieval for Sample-Based Composition
Reflector retrieves samples against the harmony of the whole evolving arrangement, not a single fixed reference clip.
The paper describes a local interactive audio workstation that tracks co-sounding regions on a multi-track timeline and updates retrieval as the session changes. Its compatibility engine starts from a hand-designed interval-class oracle, then trains a synthetic-audio encoder to approximate those scores in a 128-dimensional embedding space. The author says the learned embedding preserves the oracle’s pairwise judgments while covering the full sample library better than direct kernel scoring. Reflector is presented as free, with an open-source training pipeline and no copyrighted training data. ArXiv · AI/CL/LG's note
The paper describes a local interactive audio workstation that tracks co-sounding regions on a multi-track timeline and updates retrieval as the session changes. Its compatibility engine starts from a hand-designed interval-class oracle, then trains a synthetic-audio encoder to approximate those scores in a 128-dimensional embedding space. The author says the learned embedding preserves the oracle’s pairwise judgments while covering the full sample library better than direct kernel scoring. Reflector is presented as free, with an open-source training pipeline and no copyrighted training data. ArXiv · AI/CL/LG's note
score 4